NBER WORKING PAPER SERIES
RETIREMENT EFFECTS OF PROPOSALS BY THE
PRESIDENT’S COMMISSION TO STRENGTHEN SOCIAL SECURITY
Alan L. Gustman
Thomas L. Steinmeier
Working Paper 10030
http://www.nber.org/papers/w10030
NATIONAL BUREAU OF ECONOMIC RESEARCH
1050 Massachusetts Avenue
Cambridge, MA 02138
October 2003
Analyses of the effects of Social Security policies were supported by a grant from the U.S. Social Security
Administration (SSA) to the Michigan Retirement Research Center, with a subcontract to the National Bureau
of Economic Research. Estimation of the model on which these simulations are based was also supported by
a grant from the National Institute on Aging (1R01AG13913-01A1) to NBER. We would like to thank
Stephen Goss for his very helpful comments. The opinions and conclusions are solely those of the authors
and should not be construed as representing the opinions or policy of SSA, NIA or any other agency of the
Federal Government, the Michigan Retirement Research Center, or the National Bureau of Economic
Research. Alan L. Gustman is Loren Berry Professor of Economics at Dartmouth College, Department of
Economics, Hanover, N.H. 03755 (alan.l.gustman@dartmouth.edu). Thomas L. Steinmeier is Professor of
Economics, Texas Tech University, Department of Economics, Lubbock, Texas 79409
(Thomas.Steinmeier@TTU.edu).
©2003 by Alan L. Gustman and Thomas L. Steinmeier. All rights reserved. Short sections of text, not to
exceed two paragraphs, may be quoted without explicit permission provided that full credit, including ©
notice, is given to the source.
Retirement Effects of Proposals by the President’s Commission to Strengthen Social Security
Alan L. Gustman and Thomas L. Steinmeier
NBER Working Paper No. 10030
October 2003
JEL No. H55, J26, J14, J32, E21, D31, D91, I3
ABSTRACT
A structural dynamic model of retirement and saving is used to simulate the retirement effects of
proposals made by the President's Commission to Strengthen Social Security. Provisions reducing
the growth in real benefits and increasing actuarial incentives to work reduce retirements. They more
than offset increases in retirements caused by individual accounts, increased benefits for low wage
workers and survivors, and reductions in the top AIME bracket. By 2075, the Commission's
proposals would reduce retirements at age 62 by roughly 4 percentage points, mitigating an 8.7
percentage point trend to earlier retirement projected to reassert itself after its recent interruption.
Alan L. Gustman
Department of Economics
Dartmouth College
Hanover, NH 03755
and NBER
alan.l.gustman@dartmouth.edu
Thomas L. Steinmeier
Department of Economics
Texas Tech University
Lubbock, TX 79409
thomas.steinmeier@ttu.edu
I. Introduction
In December 2001, the President’s Commission to Strengthen Social Security (CSSS)
issued its final report. Rather than a single reform model to modify social security, the
Commission explored three different reform models, all of which include the introduction of
voluntary personal accounts. In addition, the proposals would reduce the rate of growth of real
benefits, introduce actuarial incentives to encourage later retirements, increase benefits for low
wage workers and survivors, and reduce the top AIME bracket.
The specific elements of social security that the Commission would change would also be
changed by other proposals to reform social security so as to rebalance its finances. Thus an
analysis of the retirement effects of the elements comprising the Commission’s proposal are of
more general interest than the specific models the Commission proposed. The effects we
estimate here for each specific change can be used to estimate the likely effects on retirement of
other proposals that would weight these changes differently than the President’s Commission did
in structuring its various reform proposals.
The first reform model introduced by the Commission does nothing to address the longterm funding imbalances in the social security system. The second and third proposals do
include elements to strengthen the system, and as a result these are the proposals that are being
given the most consideration.
In many ways the Commission reform models are constrained by the specific principles
given to the Commission by the President. Most importantly, the reform models are required to
include personal retirement accounts, and these must be voluntary. To make the personal
accounts attractive to the entire range of covered workers, including those who are effectively
1
receiving transfers under the current system, the Commission devised an offset system. For
every dollar of payroll taxes diverted to the personal account, future traditional benefits will be
reduced by a specified present value amount. To mitigate the solvency problem, traditional
benefits are prescribed to grow more slowly than would be the case under the current formula.
Since the current formula maintains a roughly constant replacement rate as overall earnings
increase over time, this method implies that social security will become a less important element
in the retirement resources equation.
In these proposals, the Commission also included several elements in addition to the
personal accounts. Some of these proposals would increase the relative benefits paid to lowwage workers and their widows, and others would reduce the relative benefits of higher income
workers. In one reform model, the Commission also included changes to try to induce
individuals to retire later, further reducing early retirement benefits and raising the delayed
retirement credit.
The Actuaries’ Supplement appended to the Commission report traces the effects of the
reform models on the financial health of the system and on the income levels of various groups
of individuals who would be affected by the changes. However, several of the elements of these
reform models can be expected to have non-trivial effects on retirement. These retirement
effects are not considered in the Actuaries’ Supplement or, as far as we are aware, in any other
analysis of the Commission proposals.
To simulate the retirement effects of the various elements of the Commission proposals,
we use a structural dynamic model of retirement and saving that we have developed for previous
work. This model posits lifetime expected utility that is constrained by an asset accumulation
2
equation and an uncertain lifetime. Retirement preferences and time preferences are both
allowed to be heterogeneous among workers. Workers are allowed to partially retire, usually in
a different job at a lower wage rate. Social security enters as income in the asset accumulation
equation in the years that benefits are received. The current utility value of the future benefits is,
of course, heavily dependent on the worker’s time preference rate. We apply the model to
simulate the retirement effects of the Commission proposals on a sample of married men.
In the next section, we detail more fully the aspects of the Commission proposals that we
intend to evaluate. The following section contains a brief discussion of prior efforts to model
savings and retirement. Section IV explains the model we use in the analysis and the method we
use to estimate it. A description of the simulation methodology appears in the following section.
In Section VI, we present the results of the simulations of the effects of elements of the
Commission’s proposals on retirement. A final section offers concluding remarks.
II. The CSSS Proposals
Personal accounts are the most prominent element of the Commission proposals. The
accounts themselves are fairly standard, but the offset to traditional benefits is not. For each
dollar that the individual diverts into the personal account, a dollar is placed into a fictitious
“offset” account.1 The balance in this notional offset account is compounded at a fixed real
interest rate until the individual is ready to collect benefits. At that time, the balance in the offset
account is annuitized, and the resulting amount is subtracted from the annual benefits that the
1
Gustman and Steinmeier (2002a) show that in computing the eventual retirement benefit,
the proposal both adds and subtracts the principal in the account. As a result the individual’s
benefit under the proposal is equal to the basic benefit under the relevant social security benefit
formula plus the annuitized value of any differential interest rate earnings or losses on the
personal account.
3
individual would receive under the traditional social security system. In the second Commission
reform model, the real interest rate used in the offset account is 2.0%, while in the third reform
model it is 2.5%. Both are lower than the 3.0% long-term projected real rate for treasury bonds.
If the assets in the personal accounts grow by more than the assumed offset rate, the individual
increases wealth by participation in the personal account; if not he or she loses. Since the offset
rate is lower than the rate for relatively safe treasury bonds, and since on average the varying
market rates will have decades to even themselves out, and will likely exceed the offset rates, the
expectation is that in most cases the individual would be ahead to participate in the personal
accounts.
The retirement effects of the personal accounts depend to some degree on how the payout
of the accounts is arranged. The commission’s report suggests that the withdrawals from the
accounts should begin after an individual retires, but “retirement” is a somewhat ambiguous
term. Does it mean when the individual ceases working, or when he earns less than some
minimal amount? Or does it mean when the individual works less than some number of hours in
a year? The retirement effects of the personal accounts also depend on the realizations of the
returns to those accounts. If the returns to the accounts substantially exceed expectations,
individuals might want to retire earlier, while if the accounts perform poorly individuals might
want to delay retirement. In this regard the retirement effects of the personal accounts are very
similar to the effects of holding risky assets in non-social security retirement accounts, an issue
we examined very recently in the context of the 1995-2002 stock market bubble (Gustman and
Steinmeier, 2002c). Given the findings of that study, in conjunction with the relatively small
size of the personal social security accounts, it would appear that even a major swing in stock
4
prices would probably cause no more than a percent or so change in the percentage of retired
individuals because of the personal accounts.
In this paper, we will consider the retirement effects of the personal accounts as well as
the remainder of the measures in the Commission’s reform models. Table 1 summarizes these
measures for the second and third of the Commission’s reform models. As is evident from the
table, the two reform models share some common features in addition to the personal accounts,
but there are also some important differences. In addition, the third reform model contains a
couple of features that are absent from the second reform model.
In terms of restoring the financial balance to social security, the second element listed in
the table is by far the most important. In reform model 2, the percentages in the PIA formula
would be adjusted downward every year so that the average benefit would remain roughly
constant in real terms. In reform model 3, the adjustments would hold the growth in the average
benefits to the growth in earnings less the growth in average life expectancy. In practical terms,
this means that the average real benefits in reform model 3 would grow at about 0.5 percent per
year, as opposed to a projected real earnings growth of 1 percent per year. In either case, the
replacement rate of social security benefits to earnings would be gradually reduced below levels
called for in the current formula, and the influences of social security on retirement would be
gradually reduced. The effect should be about twice as great with reform model 2 as with reform
model 3. In reform model 2, the size of traditional social security benefits, relative to wages,
would be less than half as much at the end of the 75 year planning period as they are today.
Both reform models also have provisions to maintain the benefits of long-term low-wage
workers at or above the poverty level, although the details differ. Reform model 2 would boost
5
the basic benefit of a 30 year minimum wage worker by about 40%, which is enough to provide
benefits at roughly 120 percent of the poverty level. This boost would be proportionately
reduced toward zero for workers with less than 20 years of experience and for workers with
lifetime earnings exceeding the earnings of a 35 year worker with twice the minimum wage.
Reform model 3 would give the 30 year minimum wage worker a smaller boost of 12 percent,
but the boost would extend all the way up to the average wage level before it is phased out. It
would also be phased out for individuals with less than 20 years of covered earnings, but
additional years of earnings beyond 30 would result in an even higher boost, which contrasts
with the case of reform model 2. Overall, one would expect that the boosts for low wage
workers in reform model 2 would be greater, but would cover a smaller percent of the work
force.
Reform model 3 contains a proposal to impose a larger reduction in benefits for retiring
early and to raise benefits more when retirement is delayed, so as to improve incentives to work
longer. It is the only part of any of the proposals whose primary purpose is explicitly to increase
the age at which people retire. Under current law, when the normal retirement age increases to
age 67, retirees at age 62 will receive 70 percent of full benefits, while delaying retirement past
age 67 will increase benefits by 8 percent per year. Under the proposal, age 62 retirees would
receive only 63 percent of full benefits, while delaying retirement past age 67 would increase
benefits by 10 percent per year of delay. Similar changes for spouse benefits would make the
benefits payable at age 62 only 58 percent of full benefits as opposed to 65 percent under the
current law.
A fourth element of the Commission proposals, which is essentially the same for both
6
reform models, is to increase the surviving spouse benefit of low-wage couples to 75% of the
benefit that would have been received if both spouses were still living. This provision applies if
the surviving spouse benefit is less than the average benefit for retired workers.2 Currently the
surviving spouse benefit is between 50% and 67% of the couple’s combined benefit, so the
change would raise the benefits of eligible surviving spouses by between 13 and 50 percent,
which is a nontrivial magnitude. On the other hand, at the time retirement decisions are made,
the applicability of surviving spouse benefits is probably a couple of decades away, and this
extended length of time will tend to dull the effect of this provision on retirement.
The last element listed in the table, which again applies only to reform model 3, is a
proposal to drop the percentage rate in the highest PIA bracket from 15 percent to 10 percent.
The reduction in benefits for high wage workers is much less than the one third decline in this
percentage would suggest. Even for individuals with relatively high average earnings, most of
the social security benefit comes from the first two brackets of the PIA formula, which replace
90 percent and 32 percent of average earnings, respectively. The clear intent of this proposal is
to generate some of the funds necessary to finance some of the additional subsidy given to lowwage workers, thus offsetting the negative effects of some of the other proposed changes on the
bottom part of the distribution.
The reform models have some other elements related more to the financing issue rather
than to the retirement issue. Notably, both reform models 2 and 3 call for the infusion of funds
2
In addition to its effect on the primary earner, this provision will reduce work incentives
for low wage spouses. We do not model the work behavior of spouses in this paper. For an
analysis that takes into account interdependence in the retirement decision making of husbands
and wives using the HRS, see Gustman and Steinmeier (forthcoming).
7
from the general treasury for at least some periods. It would be necessary to consider these
elements if the purpose of the paper were to examine the relative effectiveness of the reform
models to solve the solvency issue, but it is probably less critical to examine them in an analysis
of the effects of these reform models on retirement. As a result, we will limit the analyses in this
paper to the proposals listed in Table 1, which are the main proposals that can be expected to
have a significant impact on retirement.
III. Issues in Modeling the Retirement Effects of Changing Social Security Policies
There is a long tradition in modeling the effects of social security policies on retirement.
Many studies use panel data to estimate models based on a life cycle approach.3 Most of these
models can explain behavioral responses to sharp changes in the present value of rewards with
continued work. They do a poorer job of explaining responses to provisions such as the social
security early retirement age. Since benefits are reduced on an actuarially fair basis between
early and normal retirement age, there is no spike in the benefit accrual profile, and no decline in
benefit accruals to explain the sharp spike at age 62 in retirements. Thus to explain the peak in
retirements at age 62, Coile and Gruber (2002) and Panis et al. (2002) have had to include age
dummies, and the coefficient on the dummy for age 62 is strong and significant. Without
explaining this important feature of the retirement hazard, one does not have a reliable baseline
3
Much of the early literature is based on the 1969-79 Retirement History Study. See, for
example, Gordon and Blinder (1980), Burtless and Moffitt (1984), Fields and Mitchell (1984),
Gustman and Steinmeier (1985, 1986, 1991) and Rust and Phelan (1997) for contributions to this
literature based on the Retirement History Study. For a survey of the retirement literature, see
Lumsdaine and Mitchell (1999).
8
from which to judge how changes in social security rules will affect retirement outcomes.4
Gustman and Steinmeier (2002b) suggest that to explain retirement at age 62, two things
are required. First, it is necessary to relax the assumption of a perfectly operating capital market,
which characterizes all but a few structural retirement models.5 Second, it is necessary to allow
for the sharp heterogeneity in time preference which underlies the saving data, where many have
either a very low or a very high time preference rate. Without allowing for this distinction in
time preference, which our model estimates from saving data, it is not possible to explain both
the peaks in retirements, at ages 62 and 65. Nor can one explain the failure of some to respond
to the enhancements of actuarial incentives in social security such as those we analyze in this
paper.
IV. A Structural Model of Retirement and Saving
The model that we will use to simulate the effects of the Commission plans on retirement
and saving has been previously used in Gustman and Steinmeier (2002b) to estimate the effects
of raising the social security early entitlement age. Here we will sketch the nature of the model
so that readers can gain some idea of the framework of the analysis, but we refer readers back to
the earlier paper for a more detailed discussion of the model and its estimation.
The individual is assumed to maximize a utility function of consumption and leisure over
4
One can use a simple reduced form retirement model to explain the retirement peak at
age 62. All one must do is to choose a high discount rate. However, it then is not possible to
explain the peak in retirements at age 65. More generally, Gustman and Steinmeier (2000/2001)
find that the restrictions implied by a simple life cycle model with single peaked distributions of
time and leisure preferences are violated in reduced form estimates of retirement and wealth
equations. A more complex version of the life cycle model is required to jointly explain
retirement and saving behavior.
5
For exceptions, see Rust and Phalen (1997) and French (2002).
9
time:
U =
∫
2
T
0
e
−ρ t
∑{
m= 0
s m,t [ α1 C αm,t + h t γ1 L γm,t ] } dt
In this expression, C is consumption and L is the utility value of leisure. t indexes time over
the individual’s adult life, and D is the time preference rate. m indexes three survival states for
a married individual: both spouses are alive; only the respondent is alive; or only the spouse is
alive. sm,t is the probability that the couple is in the survival state denoted by m at time t.
The term ht, which is the relative value of leisure at time t, is given by the expression
h t = e X t β + ε . In this version of the model, Xt contains a constant, an age variable, an
indicator of poor health, and the birth year. The age variable presumably has a small positive
coefficient which gradually increases the value of leisure over time and reflects the gradual wear
and tear which makes the rigors of work relatively less attractive with age. As the value of
leisure increases, at some point it surpasses the utility of the consumption that continued work
makes possible, and the individual retires. It is important to note that the effects of age are
gradual, and that unlike many models of retirement, in this model there are no terms in the
preference function which would make retirement suddenly more desirable at specific ages such
as 62 or 65. The parameter , in the expression for ht varies among individuals and reflects the
fact that some individuals give leisure more weight than others. It is presumed to come from a
normal distribution with zero mean and standard deviation F,.
Individuals working more than 30 hours per week and more than 1560 hours per year are
10
classified as full-time. Those working more than 100 hours per year but less than 25 hours per
week are classified as part-time. Individuals who fall between full-time and part-time or
between part-time and retired are classified on the basis of self reports.
Leisure takes on a value of 0 for full-time work, 1 for full retirement, and ½ for
1
partial retirement. The utility value of leisure VL = γ Lγ
thus takes on a value of 0 for full-
time work, 1 for full retirement and, for partial retirement, Vp (which should take on a value Lp
between 0.5 and 1 for permissible values of (). We assume that each individual gets a random
draw of Vp from the relevant part of the exponential distribution k e
δ Vp
.
k is a constant
necessary to normalize the distribution to integrate to unity between 0.5 and 1.6 In order to
reflect that partial retirement becomes relative more common at older ages, we allow this
distribution to shift as the individual ages by specifying that * = *o + *a Age. Thus the entire
distribution of the preferences for partial retirement increases over time, although everyone
maintains their relative position in the distribution.
These preferences allow for three types of heterogeneity. The time preference parameter
D is treated as a fixed effect whose value makes the wealth implied by the model consistent with
the observable wealth in 1992 for each individual. The other two types of heterogeneity are ,,
the general preference for leisure, and Vp, the relative attractiveness of partial retirement. They
are treated as random unobserved effects within the model, coming from distributions
characterized by F,, *o and *a.
6
We initially tried a truncated normal distribution, but the relevant part of the distribution
always was estimated to be in the tail. As a result, we changed to the exponential distribution in
order to minimize the number of parameters being estimated.
11
This utility function is maximized subject to the asset constraint
At = (1 + r) At-1 + Wt (1 - Lt) + Et + Bt - Ct .
In this equation, At is the level of assets at time t, and r is the real interest rate.
The next term Wt (1 - Lt) is earnings, with the wage rate being either the full-time wage or the
partial retirement real wage rate depending on the choice of Lt. Et is the earnings of the
spouse, including any pension benefits. Bt is the sum of the individual’s own pension benefits
and family social security benefits, both of which may be influenced by the individual’s previous
and current work decisions. Ct is consumption at time t. Note that the pension and social
security amounts are actual benefits, not accruals. This means that social security and pension
wealth are not explicitly calculated but rather are implicitly determined by the fact that the
benefits enter the asset constraint in future periods.
At starts at zero in the first period, which corresponds to the year the respondent turns
age 25, and is restricted to be non-negative thereafter. Non-negativity of assets prevents
individuals with high time preference rates from borrowing against future wages, pensions, and
social security and consuming a majority of lifetime resources early in life. Rather, such
individuals will in most periods simply consume their available resources and will build up very
few assets with which to finance retirement. Individuals with lower time preferences, on the
other hand, will soon begin saving and will build up a much greater level of assets by the time
they retire.
The model is estimated for a sample of 2305 married men from the Health and
Retirement Study using observations for the first five waves of the survey, every other year from
12
1992 through 2000.7 Potential earnings profiles are taken from social security records or, if these
are not available, from the retrospective information in the respondent surveys. Future potential
earnings are projected on the basis of tenure and experience coefficients of earnings regressions.
Pension benefits, conditional on tenure in the job providing the pension, are based on
information in the summary pension descriptions, provided by the employers. Social security
benefits are based on the earnings histories and figured according to the social security rules.
The sample is limited to married men for two reasons. First, the Health and Retirement
study does not collect much information on former spouses, so it is difficult to estimate the
model for divorced or widowed men, who make up the majority of non-married men in the
sample. Secondly, it seems risky to group men and women together in the same analysis, given
the differences other researchers have found in their retirement behavior. The sample is further
reduced for several data problems. The two most important reasons for deletion, in terms of the
number of observations dropped, are for men who did not seem to be working regularly in the
pre-retirement years and men for whom the survey was not able to collect employer pension
information in the last full-time job. For the former individuals, the concept of retirement is
ambiguous if the individual does not seem to be regularly in the labor force. For the individuals
without pension information, we are potentially lacking some very important information about
retirement incentives, since pensions frequently provide strong inducements to retire at specific
dates.
The model has 8 parameters to be estimated. These include the consumption parameter
7
The Health and Retirement Study is supported principally by a grant from the National
Institute on Aging to the Institute for Social Research at the University of Michigan. Additional
support is provided by the Social Security Administration and other federal agencies.
13
", four elements of $ including the constant and coefficients for age, poor health, and birth
year, two elements of * including a constant and a coefficient of age, and the standard deviation
of retirement preferences given by F,. The model is estimated using the generalized method of
simulated moments.8 This method essentially chooses the parameters so as to minimize the
differences between a set of observed statistics (moments) in the sample and the values of those
statistics that would be implied by the model. In the minimization, the moments are weighted so
as to provide the most precise estimates possible with the data.
The estimation uses 46 moments, including the fraction of the sample working full time
and the fraction fully retired at various ages. Additional moments are calculated at various ages
for specific groups in the sample, including early and late birth cohorts, high and low lifetime
earners, and those with poor health. Only the retirement and work decisions as of the survey
dates are used in the estimates, thus avoiding the issue of having to interpolate and extrapolate
the exact ages of retirement changes between the surveys.
The results of the estimation are given in Table 2. All of the estimated parameters are
significant with the exception of the coefficient of the birth cohort variable. Perhaps the most
important parameter in these estimates is $a, the coefficient of the age variable. If this
parameter is relatively high, the value of leisure is increasing rapidly as individuals grow older,
and there is only a limited ability for economic incentives to have much of an effect on the
retirement age. If, on the other hand, the parameter is relatively low, as it is here, economic
incentives such as those contained in social security and pension rules have a much greater scope
to change retirement behavior. Another important statistic in the estimation is the q-statistic. If
8
For a description, See Greene (2000).
14
the model is correct, the q-statistic comes from a P2 distribution with the degrees of freedom
given by the number of moments minus the number of parameters. In the present model, this
translates into a P2 distribution with 38 degrees of freedom (46 moments less 8 parameters),
which has a 95% confidence bound of approximately 53.4. The estimated q-statistic of 46.9 is
well below this bound, indicating that there is nothing in the data that is inconsistent with the
model, at least using these moments.
The estimation also calculates a value of Di, the time preference rate, for each individual
in the sample. The values of Di are calculated so that for the parameter values given in Table 2,
the assets that are calculated from the model for each individual are equal to the assets (including
financial, real estate, and business assets) actually observed in 1992. The resulting distribution
of D implies a wide variation in the rates of time preference for different individuals. For about
40 percent of the individuals, the time preference rate lies in the range 0-5%, which is roughly
equal to the real interest rate. These individuals tend to have high assets relative to their income
in order to finance their retirement. At the other end of the spectrum, 27 percent have implied
time preference rates of over 50% per year, which usually means that they have accumulated
almost no assets with which to finance retirement. When they retire, their consumption will
effectively be limited to their social security and pension benefits.
V. The Simulation Procedure and the Base Simulations
The simulation procedure is straightforward. For each individual in the sample, 10,000
sets of values for the leisure preference , and the partial retirement utility Vp are drawn. Both
of these values are treated as random effects whose values are drawn from the distributions
characterized by F,, *o and *a. For each combination of the random effects, the utility value of
15
the model is evaluated for each possible age of leaving the full-time job and each possible age of
retiring fully. The individual is presumed, given the value of , and Vp, to chose the two
retirement dates so as to maximize utility.
During these evaluations, the actual potential earnings stream for the individual is used,
along with the actual pensions, health status, and other explanatory variables for the individual.
The time preference rate used in the simulations is assumed to be the rate calculated during the
estimation. In other words, the time preference rate is taken as a fixed effect. The fraction of
times that the individual retires at various ages is calculated over the 10,000 simulations for the
individual, and the results for all the individuals in the sample are aggregated to produce sample
frequencies for retirement at various ages. It should be noted that this is also the same procedure
that is used during the estimation to give the simulated moments implied by the model. For this
reason, the estimation procedure may be called the generalized method of simulated moments.
Figure 1 shows how the retirements simulated for the model compare to the actual
retirements for the sample. More properly, the retirements for the sample should probably be
called pseudo-retirements, since they are the differences between the percent retired from fulltime work at one age and the percent retired from full-time work the year before. Note that the
solid line, which represents the actual retirements from the sample, has a main peak at 62 and a
secondary peak at age 65. These two ages are, coincidentally, the early entitlement age and the
age of normal retirement in the social security system.
The simulated retirement ages, shown by the dashed line in Figure 1, reproduce these
peaks of retirement activity reasonably well. It is important to emphasize that the model
generates these peaks without introducing age dummies or splines into the utility function which
16
would induce a peak in retirement activity at these ages. This is in marked contrast to other
models, which generally have had trouble generating both peaks without using age dummies in
the utility function. The peak at age 65 is in response to the reduced actuarial adjustments as
individuals reach the normal retirement age and become subject to the delayed retirement credit
rather than the more actuarially generous early retirement penalty. For the peak at age 62, the
first thing that is likely to come to mind is that individuals with high time preference rates and no
savings need to wait until this age in order to have enough resources to support retirement, and to
some extent this is true. However, this argument is somewhat muted by the fact that many such
individuals have working spouses and/or pensions that could provide some support even if the
individual retires before 62. A perhaps more universal explanation is that for individuals with a
high discount rate, the fact that benefits must be given up in order to continue working looks like
a tax on earnings, and retirement may occur disproportionately after a reduction in net
compensation.
Figure 2 indicates the percent who are retired from full-time work and the percent who
are completely retired by age. This figure provides a reference point for interpreting the results
in the next section. For instance, suppose that a particular policy were to increase the percent
retired from full-time jobs at age 64 by 3 percentage points. The percent still working at fulltime jobs at age 64 is only about 35 percent, so such a policy would reduce the number of
individuals working at age 64 by almost 10 percent (3 percentage points vs. 35 percentage
points). One point from this diagram is that half of retirements from full-time work occur
between ages 58 and 65, so this is the age range whose retirement is most likely to be affected by
any changes in social security. A second point is that partial retirement steadily increases with
17
age, especially when compared to full-time workers. At age 58, the group of partially retired
workers is only about 10 percent as large as the pool of full-time workers, but by age 67 it
exceeds the number of full-time workers by 40 percent.
Since many of the most significant aspects of the Commission’s proposals will take effect
many years in the future, we would like to examine the effects of these changes, holding other
factors constant. To do this, we in effect move the Health and Retirement Study cohorts into the
future and do the simulations on these transplanted cohorts. For instance, consider the situation
in the years around 2050. Compared to the median cohort in the HRS, which turned 62 in 1998,
the cohort that turns 62 in 2050 would have an additional 52 years of real wage growth.
Therefore, our procedure is to take all the real earnings and pension amounts of the individuals
in the HRS and to increase them by 52 years’ worth of real wage growth, assumed to be 1
percent per year.9 Dates such as the year of birth are also set ahead by 52 years.10
Figure 3 indicates the results of these simulations. As a result of rising real wages, at age
62 retirement from full-time work would grow by over 2 percentage points by 2025, about 5
percentage points by 2050, and by over 8 percentage points by 2075, which is near the last year
of the current social security projections. This means that the fraction of individuals working
9
Source: The 2001 Annual Report of the Board of Trustees (2001, p. 82).
10
Note that we do not model other demographic changes that would characterize future
compared to current cohorts. Nor are other economic changes modeled. This implicitly assumes
that the structure of pensions relative to wages would remain the same, and similarly for the
distribution of jobs and distribution of spouse employment. By holding these other factors
constant, we isolate the effects only of changes in the level of benefits, or of other features of the
benefit structure proposed by the President’s Commission. We do, however, assume that future
cohorts will have lower mortality, consistent with the actuarial assumptions used by the Social
Security Administration.
18
full-time at age 62 would shrink by almost 15 percent by the end of this period. Note that this
increase in retirement is not coming from the cohort coefficient in the estimated utility function.
In fact the estimated coefficient is negative, which would go in the wrong direction, although the
magnitude of this coefficient is very small. Rather, the increase in retirement appears to be due
to the rising real incomes, which by 2075 would have more than doubled. Though substantial,
this increase in retirement does not seem to be unreasonably large. As a comparison, Costa
(1998; Appendix Table 2A) finds that the labor force participation rate of men age 55 to 64 fell
from 91.1 percent in 1910 to 67.0 percent in 1990.
The same pattern, though on a slightly reduced scale, appears for retirement at age 60.
At ages 65 and 67, however, the pattern is considerably different, thanks in large part to changes
in the social security rules that have already been legislated and are included in the simulations.
At age 65, the cohorts in the HRS are generally subject to the delayed retirement credit, which
for them is actuarially unfair and discourages work. By 2025, the normal retirement age will
have increased to age 67, so that individuals at age 65 will be subject to the early retirement
reduction, which is actuarially more fair and would encourage more work. This means that in
2025, full time work at age 65 should increase despite the higher lifetime incomes. After 2025,
however, retirement gradually increases by about 5 percentage points between 2025 and 2075,
which is about the same amount as at age 62. For individuals at age 67, a similar pattern
emerges. This age group is always after the normal retirement age, even in 2025. However, the
abolition of the earnings test after normal retirement age means that at that time the current
disincentives to work will be gone. The simulations suggest that the resulting increase in work
should approximately offset the increases in retirement arising from the higher lifetime income
19
levels. Again, after 2025 the increasing incomes result in a gradual increase in retirement from
full time work.
VI. Simulations of the Commission’s Proposals
In this section we will look at the retirement effects on married males of the various
elements in the two major proposals of the President’s Commission to Strengthen Social
Security. The elements we will look at are those listed in Table 1, including the personal
accounts, changes in the overall level of benefits, changes in the actuarial adjustments, special
adjustments for low wage workers, changes in the calculations of survivor benefits, and
reductions in the marginal replacement rate for those in the highest income bracket of the
primary insurance amount formula.
The most visible aspect of the proposals is to establish the personal accounts. The
contributions to the accounts are relatively clear. Under model 2, individuals would contribute
4% of covered earnings up to $1,000, while under model 3, they would contribute 2.5%. These
amounts would be carved out of the existing social security taxes. Model 3 would require an
additional 1% contribution, without any dollar limits, beyond existing taxes. Under these rules,
contributions would stop at earnings of $20,000 because of the cap, and beyond about $28,000
contributions to model 3 would be higher. Overall, total contributions to model 3 would be
expected to be higher, primarily because the additional 1% contributions do not have a cap. The
fact that model 3 has a higher interest rate on the offset account, 2.5% vs. 2%, to some extent
offsets the higher overall contribution rate in model 3.
The procedure to withdraw funds from the personal accounts, however, is not very clear.
The actuary assumes that withdrawals are converted to a joint and two-thirds variable annuity
20
when the individual retires. The problem, however, is retirement is not a very well defined
concept. It is elusive enough, in fact, that the current (or traditional) system does not use the
concept. Instead it relies on the earnings test. If an individual eligible for benefits is found
working and earning more than a certain amount, the benefits are reduced $1 for every $2 in
benefits beyond the earnings test amount, until benefits are exhausted. The lost benefits are later
restored in a more or less actuarially fair manner, but this may be overlooked by an individual
with a relatively high time preference rate. Given the elusiveness of the concept of retirement, it
is not completely clear exactly how the withdrawal of funds from the personal accounts will be
implemented.
In this paper, we will look at two possibilities. One possibility essentially mimics the
earnings test. It assumes that once an individual reaches age 62, the account is converted to a
joint and two thirds variable annuity. This annuity amount is added to the traditional benefit, and
the combined benefit is subjected to the earnings test. For every $2 that the benefits are above
the earnings test amount, $1 in benefits is lost. The lost benefits are allocated to traditional
benefits and the annuity proportional to the full amount of each type of benefit. For the part of
the traditional benefits that are lost, the usual rules for early reduction factors apply. For the part
of the annuity that is lost, that amount is converted to an annuity the next year and added to the
original annuity. For example, suppose that the annuity at age 62 is $5,000, and half of it is lost.
The lost 2,500 is converted to an annuity at age 63, with an annual amount of perhaps $125.
This would mean that the total annuity at age 63 would be the original $5,000 plus $125, or
$5,125. The other possibility we will look at in terms of the withdrawals from the personal
accounts is that individuals can convert the balances to a joint and two-thirds variable annuity,
21
and the earnings test does not apply.
In both cases, we assume that the personal accounts grow at a 4.6% real return, which the
social security actuaries calculate is the average return of a portfolio of 50% stock and 50%
bonds. Figure 4 indicates the effects of the personal annuities on retirement at age 62. In each
case, it is assumed that personal annuities are available for the entire working lives of
individuals, which is a slight overstatement for individuals reaching age 62 in 2025. In each
group, the first two bars reflect the retirement effects of the models 2 and 3, assuming that the
joint benefits will be subject to the earnings test. The last two bars in each group assume that
only the traditional benefits are subject to the earnings test. In both cases, the personal benefits
have a somewhat greater effect in model 3, probably because the amounts are higher. Also, the
effects are slightly higher when the earnings test is applied only to the traditional benefits.
Overall, it appears that the personal accounts of the size that the Commission contemplates
would raise the percentage of 62-year olds who are retired by between 1.5 and 2.5 percentage
points.
Other than the personal accounts, one of the strongest changes in the proposals is to
reduce the overall level of traditional benefits relative to the benefits provided under the current
formula. Figure 5 provides the results of simulations examining the effects of this change. In
addition to the changes in the two proposals, the figure also simulates changes to the legislated
benefits which would occur were there no other changes in the current system, so that benefits
would be lowered to match revenues when the trust fund is exhausted. These are labeled as
feasible benefits.
For each year in the figure, the left bar analyzes the effect that would occur if benefits are
22
limited to the feasible amount under current law. Benefits would hardly be affected in 2025,
since under current projections the system would still be solvent and able to pay benefits under
the current formula. Between 2025 and 2050, however, the trust fund would be exhausted, and
only about three-quarters of the currently legislated benefits could be paid. By 2075, the
percentage of benefits that could be paid is expected to drop further to about two-thirds. The
reductions in benefits make retirement less attractive, particularly for individuals with high time
preference rates. The effective reduction in compensation at age 62 from reducing benefits is
decreased for these individuals, reducing incentives to retire. By 2075 the effect of benefit level
changes on the percent retired amounts to over 4 percentage points at age 62, which means the
full-time work by 62 year-olds would increase almost 10 percent relative to what it would
otherwise be.
The second bar in each group of figures looks at the effect of holding the overall level of
benefits constant in real terms, which is the proposal in the Commission’s reform model 2. Note
that although benefits are constant in real terms, they are a shrinking fraction of real wages. As a
result, by 2075 the replacement rate of traditional social security benefits would be less than half
of the current levels. The retirement effect would grow steadily over time until, by 2075, it
would reduce retirement from full-time jobs by around 7 percentage points, which amounts to an
increase in full-time workers of around 19 percent compared to current law, or about 2.5
percentage points compared to the feasible alternative. The third bar in each group examines
indexing benefits to life expectancy, which is projected to allow real benefits to grow by about
half the rate that would occur under the current formula. This is the proposal in the
Commission’s reform model 3. Not surprisingly, the effects are roughly half of the effects of the
23
proposal to hold real benefits constant. Perhaps more surprisingly, the effects are roughly
comparable to allowing social security to run its course and, when the trust fund runs out, to pay
benefits proportional to the revenues of the system year-by-year.
The information in Figure 5 pertains to retirement from full-time work at age 62. The
effects at other ages are much as would be expected. Below age 62, before social security
benefits are available, the effects of these changes are much reduced, though not completely
trivial. After 62, the effects continue at very gradually diminishing levels as a percentage of
individuals at those ages. The diminution of the effects for older workers is undoubtedly due to
the fact that more and more of them are already retired and beyond the margin of considering
whether to delay retirement. Relative to the number of individuals still in full-time jobs,
however, the effects at age 65 are greater than those at age 62. For instance, for those age 62, the
inflation indexing proposal would increase full time work in 2075 from 37.1 percent of the
cohort to 44.1 percent, an increase of 7.0 percentage points, or 19 percent of the individuals still
working. At age 65, the corresponding percentages are 22.5 percent and 27.9 percent, which is
an increase of 5.4 percentage points but 24 percent of those still working in full-time jobs. The
net implication of these results is that these changes in the overall benefit levels would be
expected to have a substantial effect at all ages after social security benefits are available.
Similar results are obtained when we examine the effects of the various indexing schemes
on full retirement at age 62 (not shown). The magnitudes and patterns evident in Figure 5 are
reproduced with only small variation. This means that the changes in the overall level of
benefits is not merely shifting individuals from partial retirement to full-time work. Rather, they
are having the similar effects on both full-time work and work of any kind, with the implication
24
that the number of individuals in part-time work is not tremendously affected.
Figure 6 looks at the proposal to increase the penalties for early retirement. By reducing
the benefit at the early retirement age and increasing the reward to continued work through
normal retirement, incentives to work increase from the early retirement age until the normal
retirement age. The retirement effects after normal retirement age are more complex. The
model is designed so that the increase the financial rewards for continued work past the normal
retirement age depends on benefit claiming, a financial decision, but not on labor market
activity. Because the earnings test has been abolished for those over the normal retirement age,
those with a high time preference rate will no longer face an opportunity cost of not claiming
benefits when deciding how long to work past normal retirement age. They simply can claim
benefits as soon as they reach normal retirement age, and make their employment decision
independently.11 These proposals are part of the Commission’s reform model 3, but are not part
of reform model 2. They were included explicitly to provide incentives for individuals to work
longer, which presumably helps the financial situation of the trust fund. The figure considers the
effects of these provisions for various ages at various points in time. Although the changes do
not apply to the cohorts of the Health and Retirement Study, who would have by and large
already retired by the time these proposals would take effect, the first bar of each group indicates
what would have happened had they been subject to these adjustments.
Figure 6 shows substantial effects for these changes, particularly for those 65 years old.
The changes are measured as percentage points of the number of individuals at the given age.
11
The model assumes that anyone who is retired, or that those who are beyond the normal
retirement age, will claim their benefits immediately.
25
The percentage changes in those working full-time is considerably larger than the percentage
point changes. For instance, the 3.4 percentage point increase in full-time work for 65 year olds
in 2075 represents a 15 percent increase in the number of 65 year olds working full-time in that
year. The changes are in the desired direction and not inconsequential. The change for the
current cohorts at age 65 is the largest in magnitude, probably because 65 was the normal
retirement age for those cohorts, and delayed retirement credits averaging around 6 percent was
relatively low. A similar situation obtains for the current cohorts at age 67. Future cohorts
would just have reached the normal retirement age at 67, so the effects at that age for those
cohorts would be much lower.
Just focusing on future cohorts, however, the effects at age 62 and 65 are still
considerable. Between three and four percentage points of the entire cohort would be added to
the full-time employment rolls at these two ages, and at the ages in between as well. There is
some tendency for the effect to decline for the later cohorts, particularly at age 65. This is
probably due to the fact that the number of individuals in full-time work at that age will be
gradually diminishing, as indicated in Figure 3.
Figure 7 turns attention to the Commission’s proposals to offset the effects of the
proposed changes on the lowest earners. The Commission’s proposals in their reform model 2
differ somewhat from the proposals in reform model 3. Both reform models focus on longerterm workers and do not propose any changes for workers with less than 20 years of coverage.
The changes for workers with between 20 and 30 years of coverage are phased in. The provision
would boost benefits by 40 percent for minimum wage workers with 30 years of coverage under
reform model 2, and by 12 percent under reform model 3. For workers with more than 30 years
26
experience and/or less than the minimum wage, the percentage could be higher than 12 percent
in reform model 3, but not above 40 percent in reform model 2. In these calculations, the
minimum wage is presumed to grow at roughly the same rate as overall wages.
Figure 7 suggests that the effects of either of these provisions on overall full-time work
effort would be considerably lower than for either of the other two changes that have been
considered. The proposals would increase retirement, presumably because they would make
retirement more affordable for low wage workers, and the higher benefits would also increase
the penalty for continued work for those with high time preference rates. Nevertheless, neither
of these proposals would change retirement at age 62 by more than a percentage point of the
cohort size, considerably less than the several percentage point change for the previous
proposals. The effect of the proposal in reform model 2 is somewhat greater than that for reform
model 3, reflective of the fact that the benefit increases are larger in reform model 2.
Figure 8 looks at the effects of the low wage provisions on the group of respondents who
would be affected by the provisions. Among this group, the retirement effects are considerably
larger for reform model 2, reaching an increase of almost 3 percentage points in the number of
such individuals retired from full-time work. A primary reason for the larger increase in the
effect among the affected workers in reform model 2 is that the percentage of affected workers is
considerably less under reform model 2 than under reform model 3. The increased benefits for
low wage workers under reform model 2 phase out at twice the minimum wage, and only about
18.8 percent of the sample is under this limit in the HRS. The increased benefits under reform
model 3 phase out at the average earnings amount, which means that a considerably larger
percentage (38.4 percent) of the sample is affected. The smaller percentage under reform model
27
2 means that the dilution of the effect in the overall sample averages is greater than for reform
model 3.
The next figure looks at the retirement effect of increasing the survivor benefit to 75
percent of the amount that the couple would have received had both spouses survived. This
benefit increase is limited to the average primary insurance amount of all worker beneficiaries in
the previous year, reduced appropriately if the worker started taking any benefits before the
normal retirement age. Possibly because its effects are not felt until around 20 years after
retirement, the changes in the retirement probabilities are much smaller for this element of the
proposals than they have been for the previous elements. For the overall population, the
increases in retirement from full-time work at age 62 would be less than 0.3 percent for the
current generation, and around 0.1 percent for the generations approaching retirement around
2025 and afterward. Since around a third of the respondents are eligible for this benefit, the
increases for those affected are roughly three times as large. The larger effect for the current
generation relative to future generations appears connected to the fact that the future generations
will be subject to a later normal retirement age than was the case for the original HRS
generation. In any case, retirement effect of increasing survivor benefits appears to be quite
modest, especially when compared to the other changes being considered.
The final figure examines the effects of lowering the top bracket in the PIA formula to 10
percent from its current level of 15 percent. The effects of this proposal are again fairly modest,
although the effects in this case are roughly the same regardless of which generation is
considered. For all generations, the proposal would increase retirement from full-time work at
age 62 by about a quarter of a percentage point. Since not quite two-fifths of the sample would
28
be affected by this change, the effects on the part of the sample affected by the change are about
two and a half times the effect on the entire sample.
The relatively small size of changing the upper range of the PIA formula probably results
from the relatively small size of the benefit changes which would occur as a consequence. In
2000, the upper bend of the PIA formula occurred at $1,332. Looking at men who started
claiming benefits in that year, approximately 38 percent of them had PIA’s above this amount.12
The median PIA of those whose AIME was over the upper bend point appears to be between
$1,450 and $1,500 per month. Reducing the percentage applicable above the upper bend point
from 15 percent to 10 percent would lower benefits by around $50, or about 3 percent. Thus,
benefits calculated for earnings above the upper bend point constitute only a small fraction of
total benefits even for individuals above the upper bend point. As a result, it is not surprising
that this change would have only a relatively small effect on retirement.
In all of these exercises, we have not considered the retirement impact of the changes in
taxes that would necessary to finance them. According to the Actuaries’ Supplement, these costs
would range from 2.07 percent of the long-term taxable payroll for the proposal to index benefits
to inflation to 0.08 percent of the long-term taxable payroll for the proposals to increase benefits
for lower-income surviving spouses. From the participants’ point of view, the tax changes
would mainly be a wealth effect, much like the general increase in earnings over time. From the
base simulations, we saw that increasing general incomes by over 100 percent would affect
retirement at age 62 by only about 7 percentage points. Thus, the retirement effects of the taxes
12
Annual Statistical Supplement, 2001, Table 6.B4. This figure approximately agrees
with the percentage calculated from the HRS.
29
necessary to finance these changes would be tiny, implying that the changes we are finding arise
from the changes in retirement incentives. As an example, it is clear that the present system is
largely responsible for the retirement peaks at ages 62 and 65. If a proposal such as inflation
indexing benefits were to cut traditional benefits by a half over 75 years, the incentives to retire
at these ages would also be reduced by a corresponding amount.
In summary, some of the elements of the Commission’s proposals would have large
consequences for retirement, and others would have only minor consequences. Table 3 indicates
the effects of the various elements of the proposals for the retirement from full-time work for a
62 year old individual around 2075. The first line corresponds to the trend toward earlier
retirement that would occur if no changes were made. The remaining lines summarize the
effects of the various elements of the proposals, some positive and some negative. The first
column corresponds to Model 2 of the commission, and the second column corresponds to
Model 3.
In both sets of proposals, the net effect of the proposals is to reduce retirement at age 62
by about 4 percentage points. Thus, these proposals will offset about half of the trend toward
earlier retirement that overall rising incomes is likely to engender over the next 75 years. In
Model 2, the most powerful effect by far is the change to inflation indexing of benefits. This
effect alone would reduce retirement by 7 percentage points. In Model 3, an effect of the same
magnitude is split between the more moderate longevity indexing (4 percentage points) and the
increases in the actuarial rewards to continued work (3.2 percentage points). Note that Model 2
does not include the changes in the actuarial rewards. The personal accounts offset these effects
to a limited extent, about 1.5 to 2.5 percentage points. The other elements of the proposals,
30
including increasing benefits for low-wage workers, reducing benefits to high-wage workers,
and increasing survivor benefits for workers below the median wage, would produce only very
modest changes in retirement behavior.
VII. Conclusions
The model used in this paper has proven adept at explaining current patterns of
retirement behavior. By including heterogeneous time preferences and heterogenous retirement
preferences, the model is able to capture the peaks in retirement behavior at age 62 and age 65
without incorporating discontinuities in preferences which would make individuals want to retire
preferentially at those ages. The model is also able to approximate the rest of the retirement
distribution fairly accurately, and to include the non-trivial number of individuals who go
through a phase of partial retirement as well. In short, the model contains the essential elements
that permit it to analyze the effects on retirement of various potential changes which would alter
individuals’ incentives to retire, including potential changes in the social security system.
Changes in social security are very nearly a certainty, given the approaching retirement
of the baby boom. The proposals of the President’s Commission to Strengthen Social Security
are a prominent example of the potential changes that can be considered. The effects of these
proposals on the benefits of various groups and on the financial solvency of the system have
been carefully examined by the actuarial office of the system, but the potential retirement effects
of these proposals have been less well examined. The analysis presented in this paper suggests
that these effects may be substantial. Over the next 75 years, the trend toward less work and
earlier retirement, which has recently been interrupted, should continue as rising incomes induce
individuals to take a larger percentage of their potential wages as leisure. A couple of the
31
Commission’s proposals contain features which would work the other way, and would provide
individuals with incentives to delay their retirement substantially. The overall effect could offset
almost half the trend toward earlier retirement that would otherwise occur.
32
References
Burtless, Gary and Rober Moffitt. 1984. “The Effect of Social Security Benefits on the Labor
Supply of the Aged”. In H. Aaron and G. Burtless, editors. Retirement and Economic Behavior.
Brookings: Washington, D.C..
Coile, Courtney and Jonathan Gruber. 2000. “Social Security and Retirement”. NBER
Working Paper 7830.
Cost, Dora L. 1998. The Evolution of Retirement: An American Economic History, 1880-1990.
Chicago: University of Chicago Press.
Fields and Mitchell. 1984. Retirement, Pensions and Social Security. MIT Press.
French, Eric. 2002. “The Effects of Health, Wealth , and Wages on Labor Supply and
Retirement Behavior”. Xeroxed. Federal Reserve Bank of Chicago.
Greene, William H. 2000. Econometric Analysis, Fourth Edition. Upper Saddle River: Prentice
Hall.
Gordon, Roger H. and Alan S. Blinder. 1980. "Market Wages, Reservation Wages and
Retirement". Journal of Public Economics 14: 277-308.
Gustman, Alan L. and Thomas L. Steinmeier. 1985."The 1983 Social Security Reforms and
Labor Supply Adjustments of Older Individuals in the Long Run". Journal of Labor Economics,
Vol. 3, No. 2 (April, 1985) pp. 237-253.
________________. 1986. "A Structural Retirement Model." Econometrica, Vol. 54, No. 3
(May, 1986) pp. 555-584.
________________. 1991. "Changing Social Security Rules For Workers Over 65: Proposed
Policies And Their Effects", Industrial And Labor Relations Review.
________________. 2001/2002. “Retirement and Wealth”. Social Security Bulletin 64(2): 6691.
________________. 2002a. “The New Social Security Commission Personal Accounts: Where
Is the Investment Principal?” NBER Working Paper No. 9045. July, 2002.
________________. 2002b. “The Social Security Early Entitlement Age In A Structural Model
of Retirement and Wealth”. NBER Working Paper 9183. September, 2002.
________________. 2002c. “Retirement and the Stock Market Bubble”. NBER Working
Paper 9404, December.
33
________________. Forthcoming. “ Social Security, Pensions and Retirement Behavior within
the Family”. Journal of Applied Econometrics.
Lumsdaine, Robin L. and Olivia S. Mitchell. 1999. “New Developments in the Economic
Analysis of Retirement”. In Orley Ashenfelter and David Card, editors. Handbook of Labor
Economics. Amsterdam, Holland: 3261-3308.
Panis, Constantijn, Michael Hurd, David Loughran, Julie Zissimopoulos, Steven Haider and
Patricia StClair. 2002. “The Effects of Changing Social Security Administration’s Early
Retirement Age and the Normal Retirement Age.” Santa Monica: RAND.
President’s Commission to Strengthen Social Security. 2001. “Strengthening Social Security
and Creating Personal Wealth for All Americans”. Washington, D.C.: December.
http://www.commtostrengthensocsec.gov/reports/Final_report.pdf
Rust, John and Christoper Phalen. 1997. “How Social Security and Medicare Affect
Retirement”. Econometrica 65(4):781-831.
Social Security Administration. 2001. Annual Statistical Supplement 2001, to the Social
Security Bulletin. SSA Publication No. 13-11700. Washington, D.C.: U.S. Government Printing
Office.
The 2001 Annual Report of the Board of Trustees of the Federal Old-Age and Survivors
Insurance and Disability Insurance Trust Funds.
34
Table 1
Elements of the Commission Reform Models
Reform Model 2
Reform Model 3
Personal Accounts
4% contribution up to $1,000
2% real return of offset accounts
Personal Accounts
2.5% contribution up to $1,000
2.5% real return on offset accounts
1% additional contribution required
Traditional benefits held constant
in real terms
Traditional benefits indexed to
changes in life expectancy
Minimum benefit for 30 year minimum wage
worker increased to 120% of poverty level
Minimum benefit for 30 year minimum wage
worker increased to 100% of poverty level
Increased penalties for early retirement and
increased rewards for delayed retirement
Increase survivor benefit to 75%
of couple benefit
Increase survivor benefit to 75%
of couple benefit
Decrease marginal benefit for highest
AIME bracket from 15% to 10%.
35
Table 2
Estimated Results
Symbol
"
Description
Consumption parameter
Coefficient
Value
t-statistic
-0.26
8.57
-9.85
304.21
Parameters in $
$0
Constant
$a
Coefficient of Agea
0.076
5.21
$h
Coefficient of Healthd
3.92
4.57
$c
Coefficient of Cohortb,d (Year of Birth)
-0.01
0.07
-1.78
3.53
0.46
3.10
5.79
7.44
Parameters in *
*0
Constant
*a
Coefficient of Agec
F,
Standard Deviation of ,d
q value
46.92
Number of observations
2305
Several variables are differenced from their approximate means in the sample in order to
facilitate estimation. They are:
a
The actual variable is age - 62.
b
The actual variable is cohort - 1936.
c
The actual variable is age - 65.
d
These coefficients are all relative to the age coefficient, again to facilitate estimation. See
text for explanation.
36
Table 3
Effects of Proposed Social Security Changes on Retirement
from Full-Time Jobs at age 62 in 2075
Model 2
General Retirement Trend
Effects of Social Security Changes
Personal Accounts
Changes in Indexation
Changes in Actuarial Adjustments
Provisions for Low-Wage Workers
Changes in Survivor Benefits
Reduction of Top AIME Bracket
Model 3
+8.7
+1.8
-7.0
+2.7
-4.0
-3.2
+0.9
+0.1
+0.2
+0.9
+0.1
Net Effects of Social Security Changes
Overall Change in Retirement
37
+8.7
-4.2
-3.7
+4.5
+5.0
Figure 1
Retirements From Full-Time Work
16
14
Percent Retiring
12
10
8
6
4
2
0
58
59
60
61
62
63
64
Age
Retirements From FT W ork: Actual
38
Simulated
65
66
67
Figure 2
Simulated Percent Retired From Full-Time Work and Percent Fully Retired By Age
90
80
70
Percent Retired
60
50
40
30
20
10
0
58
59
60
61
62
63
64
Age
Retired from Full-Time Work
39
Fully Retired
65
66
67
Figure 3
Changes in Percent Retired from Full-Time Work Over Time by Age
10
Change in Retirement from 1998 Levels
8
6
4
2
0
60
62
65
-2
-4
Age
2025
2050
40
2075
67
Figure 4
Effects of Personal Accounts On Percent Retired From Full-Time Work at Age 62
Percentage Point Change From Outcome Under Current Law
Benefits
3
2.5
2
1.5
1
0.5
0
2025
2050
2075
Year
Model 2 with Earnings Test (ET)
Model 3 with ET
41
Model 2 without ET
Model 3 without ET
Figure 5
Effects of Adjustments in Benefit Levels On Percent Retired From Full-Time Work at Age 62
Year
2025
2050
2075
Percentage Point Change From Outcome Under Current Law
Benefits
0
-1
-2
-3
-4
-5
-6
-7
-8
Feasible Benefits
Inflation Indexing
42
Life Expectancy Indexing
Figure 6
Effects of Changes in Actuarial Adjustments on Percent Retired from Full-Time Work
Age
60
62
65
Percentage Point Change From Outcome Under Current Law
Benefits
0
-1
-2
-3
-4
-5
-6
1998
2025
43
2050
2075
67
Figure 7
Effects of Low-Wage Provisions on Percent Retired from Full-Time Work at Age 62
Percentage Point Change From Outcome Under Current Law Benefits
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
CSSS Proposal 2
CSSS Proposal 3
1998
2025
44
2050
2075
Figure 8
Effects of Low -Wage Provisions on Retirement of Low-Wage Workers
Percentage Point Change From Outcome Under Current Law Benefits
4
3.5
3
2.5
2
1.5
1
0.5
0
CSSS Proposal 2
CSSS Proposal 3
1998
2025
45
2050
2075
Figure 9
Effects of Survivor Benefit Supplements on Retirement From Full-Time Work at Age 62
Percentage Point Change From Outcome Under Current Law Benefits
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
Overall
Am ong Affected Households
1998
2025
46
2050
2075
Figure 10
Effects of Top PIA Bracket Reduction on Retirement From Full-Time Work at Age 62
Percentage Point Change From Outcome Under Current Law Benefits
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
Overall
Am ong Affected Households
1998
2025
47
2050
2075